List-enumeration capability: count entries on a WHO DON/PHEIC list (q11/q24/q25)

[smodee]

Summary

Three BFG summer-2026 questions ask the pipeline to count entries on an
authoritative list within a window, and all three get 0 usable insight
records even though the right list page is injected and survives filtering:

• q11 — How many PHEICs will be in effect on 31 Dec 2026?
• q24 — Will WHO issue a DON item for an outbreak attributed to a newly
  identified pathogen in the window? (needs to scan the DON list)
• q25 — How many new WHO Disease Outbreak News items will be published
  1 Jul–31 Dec 2026?

The insight stage extracts atomic facts from prose/tables; it has no notion of
"enumerate the entries on this list and count the ones matching a predicate
(date window / attribute)." So the injected WHO DON list and IHR/PHEIC pages
survive but yield nothing countable.

Evidence (evidence-coverage audit, live 2026-07-05)

From data/investigations/bfg_evidence_audit_2026-07-05.csv:

Q	dashboards survived	insight records	classification
q11	15	0	under_supported (cause: extraction)
q24	15	0	under_supported (cause: extraction)
q25	15 (+1 off-scope organic)	~0 usable	under_supported (cause: extraction)

These route to general_sources (the WHO DON list, IHR emergency-committee /
PHEIC page, etc.) — index/list pages, not fact pages.

What's needed

A way to enumerate list entries and count those matching a predicate, e.g.:

• q25: count DON items on who.int/emergencies/disease-outbreak-news with a
  publication date in [2026-07-01, 2026-12-31].
• q11: count active PHEIC determinations on the IHR emergency-committee page.
• q24: scan DON titles/bodies in the window for "novel pathogen" attribution.

Two plausible shapes (decide in the issue):

1. Targeted list scrapers (custom_scrapers/who_don_list.py, …) that parse
   the list, filter by date/attribute, and render a compact "N items in window"
   HTML the insight stage reads — mirrors the cdc_measles JSON-feed approach in
   BFG evidence-coverage audit + targeted extraction fixes (q16, refiner-URL, WHO H5 HAI) #53. Deterministic, no new stage.
2. An insight-stage "enumerate/count" mode — more general but larger surface.

Notes / caveats

• The WHO DON list is paginated and date-filterable via its own URL params;
  worth checking for a JSON/RSS feed first (as CDC measles had one — see BFG evidence-coverage audit + targeted extraction fixes (q16, refiner-URL, WHO H5 HAI) #53).
• This is a net-new capability, deliberately left out of the evidence-quality
  pass (BFG evidence-coverage audit + targeted extraction fixes (q16, refiner-URL, WHO H5 HAI) #53) as a recommendation.

References

• Diagnosis + write-up: data/investigations/bfg-evidence-quality-2026-07-05.md
• Evidence-quality pass: BFG evidence-coverage audit + targeted extraction fixes (q16, refiner-URL, WHO H5 HAI) #53

[smodee]

Two more questions have this exact list-enumeration shape — and one is now solved with a "targeted list scraper" (option 1), which is a useful worked example.

q19 (Chikungunya, EU/EEA) and q20 (Oropouche, Americas) both ask "how many countries/territories reported X in 2026" — i.e. enumerate a list and count. Same failure as q11/q24/q25: the injected page survives but yields nothing countable (q19 grabbed an off-scope non-EU "1"; q20 a placeholder record).

q19 is now fixed via a targeted scraper (custom_scrapers/ecdc_chikungunya.py, added in the 2026-07-05 client-side-feed pass): it routes the source to ECDC's EU/EEA seasonal report (chik-weekly.ecdc.europa.eu) and parses the summary sentence's Country (n) pairs to enumerate/count. Pre-season it renders "0 EU/EEA countries"; populated it enumerates (verified against the 2025 archive → France + Italy = 2). q19 moved to a scope-correct cumulative anchor; forecast "0"-bin 0.40 → 0.914. This is a concrete instance of option 1 working when the list lives on one authoritative page — unlike q11/q24/q25, where the list is the whole WHO DON/PHEIC index (harder, needs true list-enumeration).

q20 can't use this approach — its country list is behind a gated Tableau embed (see #50), so it needs the browser-session/snapshot path, not a list scraper.

So the enumeration set is really q11/q24/q25 + q19 (done) + q20 (blocked by #50). Ref: data/investigations/bfg-clientside-feed-tactic-2026-07-05.md.